Having NaNs in an x_in dataset leads to a jax max_steps error.

<xarray.Dataset>
Dimensions:    (id: 4, time: 9, substance: 1)
Coordinates:
  * id         (id) object 'Control' 'T 1' 'T 2' 'T3'
  * time       (time) float64 0.0 1.0 2.0 3.0 4.0 5.0 6.0 12.0 12.4
  * substance  (substance) object 'A'
Data variables:
    exposure   (id, time, substance) float64 0.0 nan nan nan ... nan 6.0 0.0 0.0

This error can be solved during Evaluator.__init__; this means the error is caught during use of dispatch_constructor

Originally posted by @flo-schu in #154